At the last decades, face analysis remains a challenging research topic in the computer vision area. Beyond the visible band, infrared images had shown several advantages for face detection and recognition. From the proposed approaches for analyzing these images, the local analysis is recognized by its feasibility to overcome typical undesirable conditions such as noise, illumination, and affine transformations. This paper describes a comprehensive study of the robustness of four of the state-of-the-art keypoint detectors i.e. SIFT, SURF, ORB and BRISK from face analysis related tasks. Robustness was evaluated regarding repeatability and precision of detected points when images are suggested to four specific transformations: blurring, noise addition, rotation, and down-scaling.